diff --git a/src/common/snippets/include/snippets/op/buffer.hpp b/src/common/snippets/include/snippets/op/buffer.hpp
index 35cd143ba4a32e..5150cdc9327fc9 100644
--- a/src/common/snippets/include/snippets/op/buffer.hpp
+++ b/src/common/snippets/include/snippets/op/buffer.hpp
@@ -13,6 +13,7 @@ namespace op {
 /**
  * @interface Buffer
  * @brief The operation is for intermediate data storage
+ * TODO
  *        - m_allocation_rank - rank of shape for memory allocation: shape[shape_rank - normalize(m_allocation_rank) : shape_rank].
  *                 It's needed to allocate needed memory size that depends on Tile rank, for example.
  *                 Default value is -1 (full shape)
@@ -29,24 +30,58 @@ class Buffer : public ngraph::op::Op {
 public:
     OPENVINO_OP("Buffer", "SnippetsOpset");
 
-    Buffer(const Output<Node>& x, const int32_t allocation_rank = -1);
-    Buffer(const ov::Shape shape, const ov::element::Type element_type, int32_t allocation_rank = -1);
+    size_t get_byte_size() const;
+    virtual ov::PartialShape get_allocation_shape() const = 0;
+
+protected:
     Buffer() = default;
+};
 
-    int32_t get_allocation_rank() const { return m_allocation_rank; }
-    void set_allocation_rank(int32_t rank) { m_allocation_rank = rank; }
+/**
+ * @interface AllocationBuffer
+ * @brief The operation is for allocation new empty memory
+ * TODO
+ * @ingroup snippets
+ */
+class AllocationBuffer : public Buffer {
+public:
+    OPENVINO_OP("AllocationBuffer", "SnippetsOpset", Buffer);
 
-    size_t get_byte_size() const;
+    AllocationBuffer() = default;
+    AllocationBuffer(const ov::Output<ov::Node>& shape, const ov::element::Type element_type);
+
+    ov::PartialShape get_allocation_shape() const override;
 
     bool visit_attributes(AttributeVisitor& visitor) override;
     std::shared_ptr<Node> clone_with_new_inputs(const OutputVector& new_args) const override;
     void validate_and_infer_types() override;
 
-private:
-    int32_t m_allocation_rank = -1;
-    ov::Shape m_static_shape;
+protected:
     ov::element::Type m_element_type;
-    bool m_is_single = false;
+};
+
+/**
+ * @interface IntermediateBuffer
+ * @brief The operation is for intermediate data storage
+ * TODO
+ * @ingroup snippets
+ */
+class IntermediateBuffer : public Buffer {
+public:
+    OPENVINO_OP("IntermediateBuffer", "SnippetsOpset", Buffer);
+
+    IntermediateBuffer() = default;
+    IntermediateBuffer(const ov::Output<ov::Node>& x);
+    IntermediateBuffer(const ov::Output<ov::Node>& x, const ov::Output<ov::Node>& shape);
+
+    ov::PartialShape get_allocation_shape() const override;
+
+    bool visit_attributes(AttributeVisitor& visitor) override { return true; }
+    std::shared_ptr<Node> clone_with_new_inputs(const OutputVector& new_args) const override;
+    void validate_and_infer_types() override;
+
+    static std::shared_ptr<ov::Node> create_shape_constant(const ov::PartialShape& shape, size_t allocation_rank);
+    static std::shared_ptr<ov::Node> create_shape_constant(const ov::PartialShape& shape);
 };
 
 } // namespace op
diff --git a/src/common/snippets/include/snippets/utils.hpp b/src/common/snippets/include/snippets/utils.hpp
index b6d4f59919d84d..16c9a284eb3a91 100644
--- a/src/common/snippets/include/snippets/utils.hpp
+++ b/src/common/snippets/include/snippets/utils.hpp
@@ -35,6 +35,10 @@ void set_output_layout(const ov::Output<Node>& port, const std::vector<size_t>&
 inline ov::Dimension get_inner_dim(const ov::PartialShape &shape) { return *(shape.rbegin()); }
 inline ov::Dimension get_outer_dim(const ov::PartialShape &shape) { return *(shape.rbegin() + 1); }
 
+inline auto normalize_rank(int32_t allocation_rank, const size_t shape_rank) -> int32_t {
+    return allocation_rank < 0 ? allocation_rank + shape_rank + 1 : allocation_rank;
+}
+
 template <typename T, typename P>
 constexpr bool one_of(T val, P item) { return val == item; }
 
diff --git a/src/common/snippets/src/op/buffer.cpp b/src/common/snippets/src/op/buffer.cpp
index 1255c13427f147..2ddbeb74cd5a44 100644
--- a/src/common/snippets/src/op/buffer.cpp
+++ b/src/common/snippets/src/op/buffer.cpp
@@ -6,8 +6,8 @@
 
 #include "snippets/op/buffer.hpp"
 #include "snippets/snippets_isa.hpp"
+#include "snippets/utils.hpp"
 
-#include <ngraph/runtime/host_tensor.hpp>
 
 using namespace std;
 using namespace ngraph;
@@ -16,62 +16,103 @@ auto normalize_rank(int32_t allocation_rank, const size_t shape_rank) -> int32_t
     return allocation_rank < 0 ? allocation_rank + static_cast<int32_t>(shape_rank) : allocation_rank;
 }
 
-snippets::op::Buffer::Buffer(const Output<Node>& x, const int32_t allocation_rank)
-    : Op({x}), m_allocation_rank(allocation_rank), m_is_single(false) {
-    constructor_validate_and_infer_types();
+size_t ngraph::snippets::op::Buffer::get_byte_size() const {
+    const auto pshape = get_allocation_shape();
+    // TODO: Add support of dynamism
+    NGRAPH_CHECK(pshape.is_static(), "Buffer should have static shapes for memory allocation");
+    const auto shape = pshape.get_shape();
+    return ngraph::shape_size(shape) * get_element_type().size();
 }
 
-snippets::op::Buffer::Buffer(const ov::Shape shape, const ov::element::Type element_type, const int32_t allocation_rank)
-    : Op(), m_static_shape(shape), m_element_type(element_type), m_allocation_rank(allocation_rank), m_is_single(true) {
+snippets::op::AllocationBuffer::AllocationBuffer(const Output<Node>& shape, const ov::element::Type element_type)
+    : Buffer(), m_element_type(element_type) {
+    set_arguments({shape});
     constructor_validate_and_infer_types();
 }
 
-bool snippets::op::Buffer::visit_attributes(AttributeVisitor& visitor) {
-    INTERNAL_OP_SCOPE(Buffer_visit_attributes);
-    visitor.on_attribute("allocation_rank", m_allocation_rank);
-    if (m_is_single) {
-        visitor.on_attribute("shape", m_static_shape);
-        visitor.on_attribute("element_type", m_element_type);
-    }
+bool snippets::op::AllocationBuffer::visit_attributes(AttributeVisitor& visitor) {
+    INTERNAL_OP_SCOPE(AllocationBuffer_visit_attributes);
+    visitor.on_attribute("element_type", m_element_type);
     return true;
 }
 
-std::shared_ptr<Node> snippets::op::Buffer::clone_with_new_inputs(const OutputVector& new_args) const {
-    INTERNAL_OP_SCOPE(Buffer_clone_with_new_inputs);
+std::shared_ptr<Node> snippets::op::AllocationBuffer::clone_with_new_inputs(const OutputVector& new_args) const {
+    INTERNAL_OP_SCOPE(AllocationBuffer_clone_with_new_inputs);
     check_new_args_count(this, new_args);
-    if (m_is_single) {
-        return std::make_shared<Buffer>(m_static_shape, m_element_type, m_allocation_rank);
+    return std::make_shared<AllocationBuffer>(new_args.at(0), m_element_type);
+}
+
+void snippets::op::AllocationBuffer::validate_and_infer_types() {
+    INTERNAL_OP_SCOPE(AllocationBuffer_validate_and_infer_types);
+    set_output_type(0, m_element_type, get_allocation_shape());
+}
+
+ov::PartialShape ngraph::snippets::op::AllocationBuffer::get_allocation_shape() const {
+    ov::PartialShape shape = ov::PartialShape::dynamic();
+    const auto shape_constant = ov::as_type_ptr<ngraph::op::v0::Constant>(get_input_node_shared_ptr(0));
+    if (shape_constant) {
+        NGRAPH_CHECK(shape_constant->get_element_type() == ov::element::i32,
+                     "The AllocationBuffer expects Constant with shape of I32 element type");
+        const auto dims = shape_constant->cast_vector<int32_t>();
+        NGRAPH_CHECK(!dims.empty(), "The AllocationBuffer got invalid shape Constant");
+        shape = ov::PartialShape(ov::Shape(std::vector<size_t>(dims.begin(), dims.end())));
     }
+    return shape;
+}
+
+snippets::op::IntermediateBuffer::IntermediateBuffer(const ov::Output<ov::Node>& x) : Buffer() {
+    set_arguments({x});
+    constructor_validate_and_infer_types();
+}
 
-    return std::make_shared<Buffer>(new_args.at(0), m_allocation_rank);
+snippets::op::IntermediateBuffer::IntermediateBuffer(const ov::Output<ov::Node>& x, const ov::Output<ov::Node>& shape) : Buffer() {
+    set_arguments({x, shape});
+    constructor_validate_and_infer_types();
 }
 
-void snippets::op::Buffer::validate_and_infer_types() {
-    INTERNAL_OP_SCOPE(Buffer_validate_and_infer_types);
-    ov::PartialShape output_shape;
-    ov::element::Type output_type;
-    if (m_is_single) {
-        output_shape = m_static_shape;
-        output_type = m_element_type;
-    } else {
-        output_shape = get_input_partial_shape(0);
-        output_type = get_input_element_type(0);
+std::shared_ptr<Node> snippets::op::IntermediateBuffer::clone_with_new_inputs(const OutputVector& new_args) const {
+    INTERNAL_OP_SCOPE(IntermediateBuffer_clone_with_new_inputs);
+    check_new_args_count(this, new_args);
+    if (new_args.size() == 2) {
+        return std::make_shared<IntermediateBuffer>(new_args.at(0), new_args.at(1));
+    } else if (new_args.size() == 1) {
+        return std::make_shared<IntermediateBuffer>(new_args.at(0));
     }
 
-    const auto shape_rank = output_shape.rank();
-    if (shape_rank.is_static()) {
-        const auto normalized_rank = normalize_rank(m_allocation_rank, shape_rank.get_length());
-        NGRAPH_CHECK(normalized_rank >= 0 && normalized_rank <= shape_rank.get_length(),
-                     "Buffer has incorrect allocation rank: " + std::to_string(m_allocation_rank));
+    throw ngraph_error("The IntermediateBuffer op got invalid input count");
+}
+
+void snippets::op::IntermediateBuffer::validate_and_infer_types() {
+    INTERNAL_OP_SCOPE(IntermediateBuffer_validate_and_infer_types);
+    set_output_type(0, get_input_element_type(0), get_input_partial_shape(0));
+}
+
+ov::PartialShape ngraph::snippets::op::IntermediateBuffer::get_allocation_shape() const {
+    if (get_input_size() == 1) {
+        return get_input_partial_shape(0);
     }
 
-    set_output_type(0, output_type, output_shape);
+    const auto shape_constant = ov::as_type_ptr<ngraph::op::v0::Constant>(get_input_node_shared_ptr(1));
+    if (shape_constant) {
+        NGRAPH_CHECK(shape_constant->get_element_type() == ov::element::i32,
+                     "The AllocationBuffer expects Constant with shape of I32 element type");
+        const auto dims = shape_constant->cast_vector<int32_t>();
+        NGRAPH_CHECK(!dims.empty(), "The AllocationBuffer got invalid shape Constant");
+        return ov::PartialShape(ov::Shape(std::vector<size_t>(dims.begin(), dims.end())));
+    }
+    return ov::PartialShape::dynamic();
 }
 
-size_t ngraph::snippets::op::Buffer::get_byte_size() const {
-    const auto pshape = get_output_partial_shape(0);
-    NGRAPH_CHECK(pshape.is_static(), "Buffer should have static shapes for memory allocation");
-    const auto shape = pshape.get_shape();
-    const auto normalized_rank = normalize_rank(m_allocation_rank, shape.size());
-    return ngraph::shape_size(shape.rbegin(), shape.rbegin() + normalized_rank + 1) * get_element_type().size();
+std::shared_ptr<ov::Node> ngraph::snippets::op::IntermediateBuffer::create_shape_constant(const ov::PartialShape& shape, size_t allocation_rank) {
+    if (shape.rank().is_dynamic())
+        return nullptr;
+    const auto normalize_rank = utils::normalize_rank(allocation_rank, shape.size());
+    const auto offset = shape.size() - normalize_rank;
+    return create_shape_constant(ov::PartialShape(std::vector<ov::Dimension>{shape.begin() + offset, shape.end()}));
 }
+
+std::shared_ptr<ov::Node> ngraph::snippets::op::IntermediateBuffer::create_shape_constant(const ov::PartialShape& shape) {
+    if (shape.is_dynamic())
+        return nullptr;
+    return std::make_shared<ov::op::v0::Constant>(ov::element::i32, ov::Shape{shape.size()}, shape.get_shape());
+}
\ No newline at end of file
diff --git a/src/common/snippets/src/op/subgraph.cpp b/src/common/snippets/src/op/subgraph.cpp
index 5e30a6d597ea02..2030a1d712cb41 100644
--- a/src/common/snippets/src/op/subgraph.cpp
+++ b/src/common/snippets/src/op/subgraph.cpp
@@ -498,6 +498,13 @@ void snippets::op::Subgraph::initialize_buffer_scratchpad_size() {
     for (const auto& op : ops) {
         if (const auto buffer = ov::as_type_ptr<ngraph::snippets::op::Buffer>(op)) {
             const auto buffer_size = buffer->get_byte_size();
+            if (ov::is_type<op::IntermediateBuffer>(op)) {
+                if (op->get_input_size() == 2) {
+                    op->set_arguments({op->get_input_source_output(0)});
+                }
+            } else if (ov::is_type<op::AllocationBuffer>(op)) {
+                op->set_arguments(ov::OutputVector{});
+            }
             // We need to allocate memory for first buffer at least
             if (m_buffer_scratchpad == 0) {
                 m_buffer_scratchpad += buffer_size;
diff --git a/src/common/snippets/src/pass/assign_registers.cpp b/src/common/snippets/src/pass/assign_registers.cpp
index 08c737c4523ae9..0805df1c5e2f33 100644
--- a/src/common/snippets/src/pass/assign_registers.cpp
+++ b/src/common/snippets/src/pass/assign_registers.cpp
@@ -38,10 +38,12 @@ bool ngraph::snippets::pass::AssignRegisters::run_on_model(const std::shared_ptr
             // here we use the fact that Result input & output tensors are identical by construction
             manually_assigned_gprs[op->output(0).get_tensor_ptr()] =
                     static_cast<Reg>(f->get_result_index(result) + num_parameters);
-        } else if (const auto& buffer = ov::as_type_ptr<op::Buffer>(op)) {
+        } else if (ov::is_type<op::Buffer>(op)) {
             // All buffers have one common data pointer
-            manually_assigned_gprs[op->input(0).get_tensor_ptr()] =
-                    static_cast<Reg>(num_results + num_parameters);
+            if (ov::is_type<op::IntermediateBuffer>(op)) {
+                manually_assigned_gprs[op->input(0).get_tensor_ptr()] =
+                        static_cast<Reg>(num_results + num_parameters);
+            }
             manually_assigned_gprs[op->output(0).get_tensor_ptr()] =
                     static_cast<Reg>(num_results + num_parameters);
         } else if (ov::is_type<op::HorizonMax>(op) || ov::is_type<op::HorizonSum>(op)) {
diff --git a/src/common/snippets/src/pass/insert_buffer.cpp b/src/common/snippets/src/pass/insert_buffer.cpp
index e3fdb0173efbb9..64b0a9ef29e77d 100644
--- a/src/common/snippets/src/pass/insert_buffer.cpp
+++ b/src/common/snippets/src/pass/insert_buffer.cpp
@@ -31,7 +31,9 @@ ngraph::snippets::pass::InsertBuffer::InsertBuffer(const int32_t allocation_rank
                 if (!ov::is_type<ngraph::snippets::op::Buffer>(input_node) &&
                     !ov::is_type<ngraph::op::v0::Parameter>(input_node) &&
                     !ov::is_type<ngraph::op::v0::Constant>(input_node)) {
-                    const auto buffer = std::make_shared<ngraph::snippets::op::Buffer>(input_node, allocation_rank);
+                    const auto constant_shape = op::IntermediateBuffer::create_shape_constant(input.get_partial_shape(), allocation_rank);
+                    const auto buffer = constant_shape ? std::make_shared<op::IntermediateBuffer>(input_node, constant_shape) :
+                                                         std::make_shared<op::IntermediateBuffer>(input_node);
                     root->set_argument(input.get_index(), buffer);
                     rewritten |= true;
                 }
@@ -67,7 +69,9 @@ ngraph::snippets::pass::InsertBuffer::InsertBuffer(const int32_t allocation_rank
                     }
                 }
 
-                const auto buffer = std::make_shared<ngraph::snippets::op::Buffer>(output, allocation_rank);
+                const auto constant_shape = op::IntermediateBuffer::create_shape_constant(output.get_partial_shape(), allocation_rank);
+                const auto buffer = constant_shape ? std::make_shared<op::IntermediateBuffer>(output, constant_shape) :
+                                    std::make_shared<op::IntermediateBuffer>(output);
                 for (const auto& consumer : output.get_target_inputs()) {
                     const auto output_node = consumer.get_node()->shared_from_this();
                     if (output_node != buffer &&
diff --git a/src/common/snippets/src/pass/insert_loops.cpp b/src/common/snippets/src/pass/insert_loops.cpp
index f26e48ec71ad07..c7afd25ce8af1d 100644
--- a/src/common/snippets/src/pass/insert_loops.cpp
+++ b/src/common/snippets/src/pass/insert_loops.cpp
@@ -137,7 +137,10 @@ void insert_loops_explicitly(const ov::NodeVector& ops, const size_t vector_size
         // on LoopBegin to guarantee that the constants are executed inside the Loop.
         for (const auto& n : body) {
             if (auto c = std::dynamic_pointer_cast<ov::op::v0::Constant>(n)) {
-                c->add_control_dependency(inner_loop_begin);
+                // Except Constant Shape for Buffers
+                if (!ov::is_type<op::Buffer>(n->get_output_target_inputs(0).begin()->get_node())) {
+                    c->add_control_dependency(inner_loop_begin);
+                }
             }
         }
 
@@ -155,6 +158,8 @@ void insert_loops_explicitly(const ov::NodeVector& ops, const size_t vector_size
             ov::is_type<ov::op::v0::Result>(op) ||
             ov::is_type<op::Buffer>(op))
             return true;
+        if (ov::is_type<ov::op::v0::Constant>(op) && ov::is_type<op::Buffer>(op->get_output_target_inputs(0).begin()->get_node()))
+            return true;
         auto& rt = op->get_rt_info();
         auto outside_rt = rt.find("outside_loop");
         bool is_outside = false;
diff --git a/src/common/snippets/src/pass/loop_fusion.cpp b/src/common/snippets/src/pass/loop_fusion.cpp
index 587daa79121ea6..cabc179c50307e 100644
--- a/src/common/snippets/src/pass/loop_fusion.cpp
+++ b/src/common/snippets/src/pass/loop_fusion.cpp
@@ -55,7 +55,7 @@ auto can_be_merged(const std::shared_ptr<ngraph::snippets::op::LoopEnd>& loop_en
 
 auto get_buffer_and_loop_end(const std::shared_ptr<ngraph::snippets::op::LoopBegin>& loop_begin_down,
                              std::shared_ptr<ngraph::snippets::op::LoopEnd>& loop_end_up,
-                             std::shared_ptr<ngraph::snippets::op::Buffer>& buffer) -> bool {
+                             std::shared_ptr<ngraph::snippets::op::IntermediateBuffer>& buffer) -> bool {
     size_t fusion_input_num = 0;
     for (const auto& parent : loop_begin_down->input_values()) {
         const auto parent_shared = parent.get_node_shared_ptr();
@@ -69,10 +69,9 @@ auto get_buffer_and_loop_end(const std::shared_ptr<ngraph::snippets::op::LoopBeg
             continue;
 
         loop_end_up = ngraph::as_type_ptr<ngraph::snippets::op::LoopEnd>(parent_shared);
-        buffer = ov::as_type_ptr<ngraph::snippets::op::Buffer>(parent_shared);
+        buffer = ov::as_type_ptr<ngraph::snippets::op::IntermediateBuffer>(parent_shared);
         if (buffer) {
             if (buffer->output(0).get_target_inputs().size() == 0 ||
-                buffer->get_input_size() != 1 ||
                 buffer->get_input_source_output(0).get_target_inputs().size() != 1)
                 return false;
 
@@ -86,7 +85,7 @@ auto get_buffer_and_loop_end(const std::shared_ptr<ngraph::snippets::op::LoopBeg
 }
 
 auto collect_loop_inputs(const std::shared_ptr<ngraph::snippets::op::LoopBegin>& loop_begin,
-                         const std::shared_ptr<ngraph::snippets::op::Buffer>& buffer,
+                         const std::shared_ptr<ngraph::snippets::op::IntermediateBuffer>& buffer,
                          std::vector<Edge>& new_loop_inputs,
                          std::vector<int64_t>& new_ptr_increments,
                          std::vector<int64_t>& new_finalization_offsets) -> void {
@@ -109,7 +108,7 @@ auto collect_loop_inputs(const std::shared_ptr<ngraph::snippets::op::LoopBegin>&
 }
 
 auto collect_loop_outputs(const std::shared_ptr<ngraph::snippets::op::LoopEnd>& loop_end,
-                          const std::shared_ptr<ngraph::snippets::op::Buffer>& buffer,
+                          const std::shared_ptr<ngraph::snippets::op::IntermediateBuffer>& buffer,
                           std::vector<Edge>& new_loop_outputs,
                           std::vector<int64_t>& new_ptr_increments,
                           std::vector<int64_t>& new_finalization_offsets,
@@ -162,7 +161,7 @@ bool ngraph::snippets::pass::LoopFusion::Merge(const std::shared_ptr<op::LoopBeg
     }
 
     std::shared_ptr<ngraph::snippets::op::LoopEnd> loop_end_up = nullptr;
-    std::shared_ptr<ngraph::snippets::op::Buffer> buffer = nullptr;
+    std::shared_ptr<ngraph::snippets::op::IntermediateBuffer> buffer = nullptr;
     // Initialize the corresponding upper LoopEnd and Buffer
     if (!get_buffer_and_loop_end(loop_begin_down, loop_end_up, buffer)) {
         return false;
diff --git a/src/common/snippets/src/pass/reset_buffer.cpp b/src/common/snippets/src/pass/reset_buffer.cpp
index bae2ac58ccdb15..55bc5aad88deae 100644
--- a/src/common/snippets/src/pass/reset_buffer.cpp
+++ b/src/common/snippets/src/pass/reset_buffer.cpp
@@ -79,10 +79,9 @@ ngraph::snippets::pass::ResetBufferState::ResetBufferState() {
 
         // If after Loop there is immediately Buffer, we should reset the Buffer ptr for the next calculations
         for (size_t i = 0; i < o_size; ++i) {
-            const auto result_shape = body_shapes[i_size + i].get_shape();
             // check for first target input is enough for Buffer searching because operations can have only single Buffer per each output port as op
             const auto consumer = loop_end->output(i).get_target_inputs().begin()->get_node();
-            if (ov::is_type<ngraph::snippets::op::Buffer>(consumer)) {
+            if (const auto buffer = ov::as_type_ptr<ngraph::snippets::op::Buffer>(consumer->shared_from_this())) {
                 // To calculate finalization offset we should know index of nesting Loop
                 auto loop_index = 0lu;
                 auto loop = loop_end->input_value(i).get_node_shared_ptr();
@@ -93,7 +92,10 @@ ngraph::snippets::pass::ResetBufferState::ResetBufferState() {
                     port_idx = source_output.get_index();
                     loop_index++;
                 }
-
+                const auto pshape = buffer->get_allocation_shape();
+                NGRAPH_CHECK(pshape.is_static(), "Buffer must have static allocation shape to calculate finalization offsets");
+                const auto result_shape = pshape.get_shape();
+                NGRAPH_CHECK(loop_index < result_shape.size(), "Buffer has invalid Loop index and allocation shape rank");
                 const auto work_amount = std::accumulate(result_shape.rbegin(), result_shape.rbegin() + loop_index + 1, size_t(1), std::multiplies<size_t>());
                 finalization_offsets[i_size + i] =
                         calculate_required_finalization_offsets(work_amount, *(result_shape.rbegin() + loop_index));
diff --git a/src/common/snippets/src/pass/softmax_decomposition.cpp b/src/common/snippets/src/pass/softmax_decomposition.cpp
index 1a7330fb537641..0e3713ebdf6a79 100644
--- a/src/common/snippets/src/pass/softmax_decomposition.cpp
+++ b/src/common/snippets/src/pass/softmax_decomposition.cpp
@@ -125,7 +125,9 @@ ngraph::snippets::pass::SoftmaxDecomposition::SoftmaxDecomposition(const size_t
             apply_increments_sum, finalization_offsets_sum);
 
         const auto horizon_sum = std::make_shared<ngraph::snippets::op::HorizonSum>(sum);
-        const auto buffer_exp = std::make_shared<ngraph::snippets::op::Buffer>(loop_sum_end->output(0), buffer_allocation_rank);
+        const auto constant_shape_exp = op::IntermediateBuffer::create_shape_constant(loop_sum_end->output(0).get_partial_shape(), buffer_allocation_rank);
+        const auto buffer_exp = constant_shape_exp ? std::make_shared<op::IntermediateBuffer>(loop_sum_end->output(0), constant_shape_exp) :
+                            std::make_shared<op::IntermediateBuffer>(loop_sum_end->output(0));
 
         /* =========================================== */
 
diff --git a/src/common/snippets/tests/src/lowering_utils.cpp b/src/common/snippets/tests/src/lowering_utils.cpp
index a07fb4c0884dbe..c557a62c75e866 100644
--- a/src/common/snippets/tests/src/lowering_utils.cpp
+++ b/src/common/snippets/tests/src/lowering_utils.cpp
@@ -38,7 +38,8 @@ DummyTargetMachine::DummyTargetMachine() {
     jitters[ngraph::snippets::op::LoopBegin::get_type_info_static()] = dummy_functor;
     jitters[ngraph::snippets::op::LoopEnd::get_type_info_static()] = dummy_functor;
     jitters[ngraph::snippets::op::Brgemm::get_type_info_static()] = dummy_functor;
-    jitters[ngraph::snippets::op::Buffer::get_type_info_static()] = dummy_functor;
+    jitters[ngraph::snippets::op::IntermediateBuffer::get_type_info_static()] = dummy_functor;
+    jitters[ngraph::snippets::op::AllocationBuffer::get_type_info_static()] = dummy_functor;
     jitters[ngraph::snippets::op::VectorBuffer::get_type_info_static()] = dummy_functor;
     jitters[ngraph::snippets::op::Fill::get_type_info_static()] = dummy_functor;
 }
diff --git a/src/common/snippets/tests/src/pass/merge_loops.cpp b/src/common/snippets/tests/src/pass/merge_loops.cpp
index be398f2107fd9a..1a51a0365b17d1 100644
--- a/src/common/snippets/tests/src/pass/merge_loops.cpp
+++ b/src/common/snippets/tests/src/pass/merge_loops.cpp
@@ -38,7 +38,7 @@ TEST(TransformationTests, UnaryEltwisesLoops) {
                 OutputVector{inner_loop_end_up->output(0), outer_loop_begin_up->output(1)}, shape[shape.size() - 2], 1,
                 std::vector<int64_t>{0, 0}, std::vector<int64_t>{0, 0});
 
-        auto buffer = std::make_shared<snippets::op::Buffer>(outer_loop_end_up);
+        auto buffer = std::make_shared<snippets::op::IntermediateBuffer>(outer_loop_end_up);
 
         auto outer_loop_begin_down = std::make_shared<snippets::op::LoopBegin>(OutputVector{buffer});
         auto inner_loop_begin_down = std::make_shared<snippets::op::LoopBegin>(OutputVector{outer_loop_begin_down});
@@ -108,7 +108,7 @@ TEST(TransformationTests, BinaryEltwisesLoops) {
                 OutputVector{inner_loop_end_up->output(0), outer_loop_begin_up->output(2)}, shape[shape.size() - 2], 1,
                 std::vector<int64_t>{0, 0, 0}, std::vector<int64_t>{0, 0, 0});
 
-        auto buffer = std::make_shared<snippets::op::Buffer>(outer_loop_end_up);
+        auto buffer = std::make_shared<snippets::op::IntermediateBuffer>(outer_loop_end_up);
 
         auto data2 = std::make_shared<opset1::Parameter>(element::f32, shape);
 
diff --git a/src/plugins/intel_cpu/src/emitters/cpu_generator.cpp b/src/plugins/intel_cpu/src/emitters/cpu_generator.cpp
index 66fb1f56537bf3..63d30e0a65223e 100644
--- a/src/plugins/intel_cpu/src/emitters/cpu_generator.cpp
+++ b/src/plugins/intel_cpu/src/emitters/cpu_generator.cpp
@@ -47,7 +47,8 @@ ov::intel_cpu::CPUTargetMachine::CPUTargetMachine(dnnl::impl::cpu::x64::cpu_isa_
     // data movement
     jitters[ngraph::opset1::Parameter::get_type_info_static()] = CREATE_EMITTER(NopEmitter);
     jitters[ngraph::opset1::Result::get_type_info_static()] = CREATE_EMITTER(NopEmitter);
-    jitters[ngraph::snippets::op::Buffer::get_type_info_static()] = CREATE_EMITTER(NopEmitter);
+    jitters[ngraph::snippets::op::AllocationBuffer::get_type_info_static()] = CREATE_EMITTER(NopEmitter);
+    jitters[ngraph::snippets::op::IntermediateBuffer::get_type_info_static()] = CREATE_EMITTER(NopEmitter);
     jitters[ngraph::snippets::op::VectorBuffer::get_type_info_static()] = CREATE_EMITTER(VectorBufferEmitter);
     // jitters[ngraph::opset1::Constant::get_type_info_static()] = CREATE_EMITTER(); // Not supported
 
diff --git a/src/plugins/intel_cpu/src/extension.cpp b/src/plugins/intel_cpu/src/extension.cpp
index f1a0d68c8dbaaa..43d54e32ead243 100644
--- a/src/plugins/intel_cpu/src/extension.cpp
+++ b/src/plugins/intel_cpu/src/extension.cpp
@@ -135,15 +135,16 @@ std::map<std::string, ngraph::OpSet> Extension::getOpSets() {
         ngraph::OpSet opset;
 
 #define NGRAPH_OP(NAME, NAMESPACE) opset.insert<NAMESPACE::NAME>();
+        NGRAPH_OP(AllocationBuffer, ngraph::snippets::op)
         NGRAPH_OP(Brgemm, ngraph::snippets::op)
         NGRAPH_OP(BroadcastLoad, ngraph::snippets::op)
         NGRAPH_OP(BroadcastMove, ngraph::snippets::op)
-        NGRAPH_OP(Buffer, ngraph::snippets::op)
         NGRAPH_OP(ConvertSaturation, ngraph::snippets::op)
         NGRAPH_OP(ConvertTruncation, ngraph::snippets::op)
         NGRAPH_OP(Fill, ngraph::snippets::op)
         NGRAPH_OP(HorizonMax, ngraph::snippets::op)
         NGRAPH_OP(HorizonSum, ngraph::snippets::op)
+        NGRAPH_OP(IntermediateBuffer, ngraph::snippets::op)
         NGRAPH_OP(Kernel, ngraph::snippets::op)
         NGRAPH_OP(Load, ngraph::snippets::op)
         NGRAPH_OP(LoadReshape, ngraph::snippets::op)
diff --git a/src/plugins/intel_cpu/src/snippets_transformations/brgemm_to_brgemm_cpu.cpp b/src/plugins/intel_cpu/src/snippets_transformations/brgemm_to_brgemm_cpu.cpp
index f7dc571fce82e8..b9305748ebbe55 100644
--- a/src/plugins/intel_cpu/src/snippets_transformations/brgemm_to_brgemm_cpu.cpp
+++ b/src/plugins/intel_cpu/src/snippets_transformations/brgemm_to_brgemm_cpu.cpp
@@ -65,12 +65,17 @@ pass::BrgemmToBrgemmCPU::BrgemmToBrgemmCPU() {
         } else {
             const auto layoutIn1 = ngraph::snippets::utils::get_node_output_layout(brgemm->input_value(1).get_node_shared_ptr());
             const auto brgemmRepackIn1 = std::make_shared<BrgemmCopyB>(brgemm->input_value(1), element_type_a, with_comp, offset_b);
-            const auto buffer = std::make_shared<ngraph::snippets::op::Buffer>(brgemmRepackIn1->output(0));
-            const auto scratch = with_amx ? std::make_shared<ngraph::snippets::op::Buffer>(ov::Shape{4 * 1024}, ov::element::i32) :
-                                 with_comp ? std::make_shared<ngraph::snippets::op::Buffer>(brgemmRepackIn1->output(1)) :
-                                         nullptr;
+            const auto buffer = std::make_shared<ngraph::snippets::op::IntermediateBuffer>(brgemmRepackIn1->output(0));
 
             if (with_amx || with_comp) {
+                std::shared_ptr<ngraph::snippets::op::Buffer> scratch = nullptr;
+                if (with_amx) {
+                    const auto scratch_size = std::make_shared<ov::op::v0::Constant>(ov::element::i32, ov::Shape{1}, std::vector<int32_t>{4 * 1024});
+                    scratch = std::make_shared<ngraph::snippets::op::AllocationBuffer>(scratch_size, ov::element::i32);
+                } else if (with_comp) {
+                    scratch = std::make_shared<ngraph::snippets::op::IntermediateBuffer>(brgemmRepackIn1->output(1));
+                }
+
                 brgemm_cpu = std::make_shared<BrgemmCPU>(brgemm->input_value(0), buffer, scratch,
                                                          brgemm->transposed_a(), brgemm->transposed_b(), with_comp,
                                                          offset_a, offset_b, offset_c);
diff --git a/src/plugins/intel_cpu/src/snippets_transformations/op/brgemm_cpu.cpp b/src/plugins/intel_cpu/src/snippets_transformations/op/brgemm_cpu.cpp
index 918a66f58c7bbe..df8df656f694d6 100644
--- a/src/plugins/intel_cpu/src/snippets_transformations/op/brgemm_cpu.cpp
+++ b/src/plugins/intel_cpu/src/snippets_transformations/op/brgemm_cpu.cpp
@@ -113,7 +113,7 @@ std::shared_ptr<Node> BrgemmCPU::clone_with_new_inputs(const OutputVector& new_a
 }
 
 std::shared_ptr<BrgemmCopyB> BrgemmCPU::get_brgemm_copy() const {
-    if (const auto buffer = ov::as_type_ptr<ngraph::snippets::op::Buffer>(get_input_node_shared_ptr(1))) {
+    if (const auto buffer = ov::as_type_ptr<ngraph::snippets::op::IntermediateBuffer>(get_input_node_shared_ptr(1))) {
         return ov::as_type_ptr<BrgemmCopyB>(buffer->get_input_node_shared_ptr(0));
     }
     return nullptr;
diff --git a/src/tests/ngraph_helpers/snippets_ngraph_functions/src/subgraph_lowered.cpp b/src/tests/ngraph_helpers/snippets_ngraph_functions/src/subgraph_lowered.cpp
index 31e66c97534164..c1cf7e2db01167 100644
--- a/src/tests/ngraph_helpers/snippets_ngraph_functions/src/subgraph_lowered.cpp
+++ b/src/tests/ngraph_helpers/snippets_ngraph_functions/src/subgraph_lowered.cpp
@@ -194,7 +194,7 @@ std::shared_ptr<ov::Model> SoftmaxLoweredFunction::initLowered() const {
     const auto horizon_sum = std::make_shared<ngraph::snippets::op::HorizonSum>(sum);
     horizon_sum->add_control_dependency(loop_sum_end);
 
-    const auto buffer_exp = std::make_shared<ngraph::snippets::op::Buffer>(loop_sum_end->output(0));
+    const auto buffer_exp = std::make_shared<ngraph::snippets::op::IntermediateBuffer>(loop_sum_end->output(0));
 
     loop_sum_begin->add_control_dependency(vector_buffer_sum);
     loop_sum_begin->add_control_dependency(horizon_max);
@@ -303,7 +303,7 @@ std::shared_ptr<ov::Model> AddSoftmaxLoweredFunction::initLowered() const {
 
     /* =========================================== */
 
-    const auto buffer_add = std::make_shared<ngraph::snippets::op::Buffer>(loop_max_end->output(0));
+    const auto buffer_add = std::make_shared<ngraph::snippets::op::IntermediateBuffer>(loop_max_end->output(0));
 
     /* === Sub + Exp + ReduceSum decomposition === */
 
@@ -331,7 +331,7 @@ std::shared_ptr<ov::Model> AddSoftmaxLoweredFunction::initLowered() const {
     const auto horizon_sum = std::make_shared<ngraph::snippets::op::HorizonSum>(sum);
     horizon_sum->add_control_dependency(loop_sum_end);
 
-    const auto buffer_exp = std::make_shared<ngraph::snippets::op::Buffer>(loop_sum_end->output(0));
+    const auto buffer_exp = std::make_shared<ngraph::snippets::op::IntermediateBuffer>(loop_sum_end->output(0));
 
     loop_sum_begin->add_control_dependency(vector_buffer_sum);
     loop_sum_begin->add_control_dependency(horizon_max);