diff --git a/src/plugins/intel_gpu/src/graph/impls/cpu/detection_output.cpp b/src/plugins/intel_gpu/src/graph/impls/cpu/detection_output.cpp
index d211b610c0c3d7..fac3a6428734a9 100644
--- a/src/plugins/intel_gpu/src/graph/impls/cpu/detection_output.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/cpu/detection_output.cpp
@@ -68,11 +68,11 @@ struct detection_output_impl : typed_primitive_impl<detection_output> {
         nms_type = (node.get_primitive()->decrease_label_id ? NMSType::MXNET : NMSType::CAFFE);
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
+    void save(BinaryOutputBuffer& ob) const override {
         ob << make_data(&nms_type, sizeof(NMSType));
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
+    void load(BinaryInputBuffer& ib) override {
         ib >> make_data(&nms_type, sizeof(NMSType));
     }
 
diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/activation.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/activation.cpp
index 6b367a00b14d71..c3bb3bac00f000 100644
--- a/src/plugins/intel_gpu/src/graph/impls/ocl/activation.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/ocl/activation.cpp
@@ -48,13 +48,13 @@ struct activation_impl : typed_primitive_impl_ocl<activation> {
         return args;
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _is_parameterized;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _is_parameterized;
     }
 
diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/binary_convolution.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/binary_convolution.cpp
index 096aa56db787da..7d851c891643ad 100644
--- a/src/plugins/intel_gpu/src/graph/impls/ocl/binary_convolution.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/ocl/binary_convolution.cpp
@@ -75,13 +75,13 @@ struct binary_convolution_impl : typed_primitive_impl_ocl<binary_convolution> {
     int32_t get_split() const override { return _split; }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _split;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _split;
     }
 
diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/concatenation.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/concatenation.cpp
index 5286df10e71868..54a8caa445083d 100644
--- a/src/plugins/intel_gpu/src/graph/impls/ocl/concatenation.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/ocl/concatenation.cpp
@@ -85,13 +85,13 @@ struct concatenation_impl : typed_primitive_impl_ocl<concatenation> {
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _can_be_optimized;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _can_be_optimized;
     }
 
diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/convolution.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/convolution.cpp
index e0110774718bbb..aa7a918aaa1028 100644
--- a/src/plugins/intel_gpu/src/graph/impls/ocl/convolution.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/ocl/convolution.cpp
@@ -80,15 +80,15 @@ struct convolution_impl : typed_primitive_impl_ocl<convolution> {
     bool get_depthwise_sep_opt() const override { return _depthwise_sep_opt; }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _split;
         ob << _groups;
         ob << _depthwise_sep_opt;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _split;
         ib >> _groups;
         ib >> _depthwise_sep_opt;
diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/crop.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/crop.cpp
index 4807a8ddf4634a..0d861a61fbcb6d 100644
--- a/src/plugins/intel_gpu/src/graph/impls/ocl/crop.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/ocl/crop.cpp
@@ -44,13 +44,13 @@ struct crop_impl : typed_primitive_impl_ocl<crop> {
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _can_be_optimized;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _can_be_optimized;
     }
 
diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/custom_primitive.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/custom_primitive.cpp
index f384e70797c210..3421e8c5718b07 100644
--- a/src/plugins/intel_gpu/src/graph/impls/ocl/custom_primitive.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/ocl/custom_primitive.cpp
@@ -85,12 +85,12 @@ struct custom_gpu_primitive_impl : typed_primitive_impl<custom_gpu_primitive> {
         return {_kernel_id};
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
+    void save(BinaryOutputBuffer& ob) const override {
         ob << *cl_kernel;
         ob << _kernel_id;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
+    void load(BinaryInputBuffer& ib) override {
         cl_kernel = std::make_shared<kernel_selector::cl_kernel_data>();
         ib >> *cl_kernel;
         ib >> _kernel_id;
diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/deconvolution.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/deconvolution.cpp
index e214b33962a17f..7d939b66288e70 100644
--- a/src/plugins/intel_gpu/src/graph/impls/ocl/deconvolution.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/ocl/deconvolution.cpp
@@ -41,14 +41,14 @@ struct deconvolution_impl : typed_primitive_impl_ocl<deconvolution> {
         _groups = node.get_groups();
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _split;
         ob << _groups;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _split;
         ib >> _groups;
     }
diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/deformable_convolution.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/deformable_convolution.cpp
index 50fa6ca5f4d2c8..57255cc12c3468 100644
--- a/src/plugins/intel_gpu/src/graph/impls/ocl/deformable_convolution.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/ocl/deformable_convolution.cpp
@@ -42,14 +42,14 @@ struct deformable_conv_impl : typed_primitive_impl_ocl<deformable_conv> {
         _groups = node.get_groups();
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _split;
         ob << _groups;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _split;
         ib >> _groups;
     }
diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/generic_layer.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/generic_layer.cpp
index 7283b3f56c5128..a25203538dd00e 100644
--- a/src/plugins/intel_gpu/src/graph/impls/ocl/generic_layer.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/ocl/generic_layer.cpp
@@ -44,12 +44,12 @@ struct generic_layer_impl : typed_primitive_impl<generic_layer> {
         _kernel_id = arg.get_program().add_kernel(arg.get_primitive()->generic_params.clKernel->code.kernelString);
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
+    void save(BinaryOutputBuffer& ob) const override {
         ob <<_cl_kernel_data;
         ob << _kernel_id;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
+    void load(BinaryInputBuffer& ib) override {
         ib >> _cl_kernel_data;
         ib >> _kernel_id;
     }
diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/primitive_base.hpp b/src/plugins/intel_gpu/src/graph/impls/ocl/primitive_base.hpp
index 05711264aa017d..ee4fd671733ab3 100644
--- a/src/plugins/intel_gpu/src/graph/impls/ocl/primitive_base.hpp
+++ b/src/plugins/intel_gpu/src/graph/impls/ocl/primitive_base.hpp
@@ -73,7 +73,7 @@ struct typed_primitive_impl_ocl : public typed_primitive_impl<PType> {
 
     bool is_cpu() const override { return false; }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
+    void save(BinaryOutputBuffer& ob) const override {
         ob << make_data(&_kernel_data.internalBufferDataType, sizeof(kernel_selector::Datatype));
         ob << _kernel_data.internalBufferSizes;
         ob << _kernel_data.kernels;
@@ -81,7 +81,7 @@ struct typed_primitive_impl_ocl : public typed_primitive_impl<PType> {
         ob << _kernel_args;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
+    void load(BinaryInputBuffer& ib) override {
         ib >> make_data(&_kernel_data.internalBufferDataType, sizeof(kernel_selector::Datatype));
         ib >> _kernel_data.internalBufferSizes;
         ib >> _kernel_data.kernels;
diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/reorder.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/reorder.cpp
index c5b24f5a7ae7de..0e6c6f8f24ba10 100644
--- a/src/plugins/intel_gpu/src/graph/impls/ocl/reorder.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/ocl/reorder.cpp
@@ -40,14 +40,14 @@ struct reorder_impl : typed_primitive_impl_ocl<reorder> {
         _has_mean = node.has_mean();
     }
 
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
         ob << _can_be_optimized;
         ob << _has_mean;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
         ib >> _can_be_optimized;
         ib >> _has_mean;
     }
diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/concatenation_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/concatenation_onednn.cpp
index 3c6bbeac5d71f0..bf89db82a19fde 100644
--- a/src/plugins/intel_gpu/src/graph/impls/onednn/concatenation_onednn.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/onednn/concatenation_onednn.cpp
@@ -63,7 +63,7 @@ struct concatenation_onednn : typed_primitive_onednn_impl<concatenation, void, d
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
+    void save(BinaryOutputBuffer& ob) const override {
         if (_prim.get(true) == nullptr) {
             ob << false;
             return;
@@ -71,8 +71,9 @@ struct concatenation_onednn : typed_primitive_onednn_impl<concatenation, void, d
             ob << true;
         }
 
-        parent::save(ob, impl_params);
+        parent::save(ob);
 
+        const kernel_impl_params* impl_params = reinterpret_cast<kernel_impl_params*>(ob.getKernlImplParams());
         auto prim = impl_params->typed_desc<concatenation>();
         ob << prim->axis;
 
@@ -81,18 +82,19 @@ struct concatenation_onednn : typed_primitive_onednn_impl<concatenation, void, d
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
+    void load(BinaryInputBuffer& ib) override {
         bool has_prim;
         ib >> has_prim;
 
         if (!has_prim)
             return;
 
-        parent::load(ib, impl_params);
+        parent::load(ib);
 
         int64_t prim_axis;
         ib >> prim_axis;
 
+        const kernel_impl_params* impl_params = reinterpret_cast<kernel_impl_params*>(ib.getKernlImplParams());
         auto desc = get_concatenation_descriptor(*impl_params, prim_axis, ib.get_engine());
         _pd = *desc;
 
diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/convolution_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/convolution_onednn.cpp
index 27c1f2f94f8f01..f10dde97ceaa96 100644
--- a/src/plugins/intel_gpu/src/graph/impls/onednn/convolution_onednn.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/onednn/convolution_onednn.cpp
@@ -185,8 +185,8 @@ struct convolution_onednn : typed_primitive_onednn_impl<convolution, dnnl::convo
 
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         ob << make_data(&_desc->data, sizeof(dnnl_convolution_desc_t));
 
@@ -195,8 +195,8 @@ struct convolution_onednn : typed_primitive_onednn_impl<convolution, dnnl::convo
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
         _desc = std::make_shared<dnnl::convolution_forward::desc>();
         ib >> make_data(&_desc->data, sizeof(dnnl_convolution_desc_t));
diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/deconvolution_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/deconvolution_onednn.cpp
index cf40323f8d1528..8be3740fbdff54 100644
--- a/src/plugins/intel_gpu/src/graph/impls/onednn/deconvolution_onednn.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/onednn/deconvolution_onednn.cpp
@@ -104,8 +104,8 @@ struct deconvolution_onednn : typed_primitive_onednn_impl<deconvolution, dnnl::d
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         ob << make_data(&_desc->data, sizeof(dnnl_deconvolution_desc_t));
 
@@ -114,8 +114,8 @@ struct deconvolution_onednn : typed_primitive_onednn_impl<deconvolution, dnnl::d
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
         _desc = std::make_shared<dnnl::deconvolution_forward::desc>();
         ib >> make_data(&_desc->data, sizeof(dnnl_deconvolution_desc_t));
diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/fully_connected_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/fully_connected_onednn.cpp
index 99b5aea13bc91a..af875b5f6b38b9 100644
--- a/src/plugins/intel_gpu/src/graph/impls/onednn/fully_connected_onednn.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/onednn/fully_connected_onednn.cpp
@@ -166,8 +166,8 @@ struct fully_connected_onednn : typed_primitive_onednn_impl<fully_connected, dnn
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         ob << make_data(&_desc->data, sizeof(dnnl_inner_product_desc_t));
 
@@ -176,8 +176,8 @@ struct fully_connected_onednn : typed_primitive_onednn_impl<fully_connected, dnn
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
         _desc = std::make_shared<dnnl::inner_product_forward::desc>();
         ib >> make_data(&_desc->data, sizeof(dnnl_inner_product_desc_t));
diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/gemm_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/gemm_onednn.cpp
index 20f8ced65186a1..126c0c59b9c893 100644
--- a/src/plugins/intel_gpu/src/graph/impls/onednn/gemm_onednn.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/onednn/gemm_onednn.cpp
@@ -209,8 +209,8 @@ struct gemm_onednn : typed_primitive_onednn_impl<gemm, dnnl::matmul::desc> {
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         ob << make_data(&_desc->data, sizeof(dnnl_matmul_desc_t));
 
@@ -219,8 +219,8 @@ struct gemm_onednn : typed_primitive_onednn_impl<gemm, dnnl::matmul::desc> {
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
         _desc = std::make_shared<dnnl::matmul::desc>();
         ib >> make_data(&_desc->data, sizeof(dnnl_matmul_desc_t));
diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/pooling_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/pooling_onednn.cpp
index 20eae95f89c00b..6707da450898d0 100644
--- a/src/plugins/intel_gpu/src/graph/impls/onednn/pooling_onednn.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/onednn/pooling_onednn.cpp
@@ -64,8 +64,8 @@ struct pooling_onednn : typed_primitive_onednn_impl<pooling, dnnl::pooling_forwa
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         ob << make_data(&_desc->data, sizeof(dnnl_pooling_desc_t));
 
@@ -74,8 +74,8 @@ struct pooling_onednn : typed_primitive_onednn_impl<pooling, dnnl::pooling_forwa
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
         _desc = std::make_shared<dnnl::pooling_forward::desc>();
         ib >> make_data(&_desc->data, sizeof(dnnl_pooling_desc_t));
diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/reduction_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/reduction_onednn.cpp
index e4a6834183882d..1f25295e4b3fac 100644
--- a/src/plugins/intel_gpu/src/graph/impls/onednn/reduction_onednn.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/onednn/reduction_onednn.cpp
@@ -93,8 +93,8 @@ struct reduction_onednn : typed_primitive_onednn_impl<reduce, dnnl::reduction::d
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         ob << make_data(&_desc->data, sizeof(dnnl_reduction_desc_t));
 
@@ -103,8 +103,8 @@ struct reduction_onednn : typed_primitive_onednn_impl<reduce, dnnl::reduction::d
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
         _desc = std::make_shared<dnnl::reduction::desc>();
         ib >> make_data(&_desc->data, sizeof(dnnl_reduction_desc_t));
diff --git a/src/plugins/intel_gpu/src/graph/impls/onednn/reorder_onednn.cpp b/src/plugins/intel_gpu/src/graph/impls/onednn/reorder_onednn.cpp
index 1575e388088287..bfe776d9a64b46 100644
--- a/src/plugins/intel_gpu/src/graph/impls/onednn/reorder_onednn.cpp
+++ b/src/plugins/intel_gpu/src/graph/impls/onednn/reorder_onednn.cpp
@@ -63,17 +63,18 @@ struct reorder_onednn : typed_primitive_onednn_impl<reorder, void, dnnl::reorder
     }
 
 public:
-    void save(BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const override {
-        parent::save(ob, impl_params);
+    void save(BinaryOutputBuffer& ob) const override {
+        parent::save(ob);
 
         std::vector<uint8_t> prim_cache;
         prim_cache = _prim.get_cache_blob();
         ob << prim_cache;
     }
 
-    void load(BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) override {
-        parent::load(ib, impl_params);
+    void load(BinaryInputBuffer& ib) override {
+        parent::load(ib);
 
+        const kernel_impl_params* impl_params = reinterpret_cast<kernel_impl_params*>(ib.getKernlImplParams());
         auto desc = get_reorder_descriptor(*impl_params, *_attrs, ib.get_engine());
         _pd = *desc;
 
diff --git a/src/plugins/intel_gpu/src/graph/include/primitive_inst.h b/src/plugins/intel_gpu/src/graph/include/primitive_inst.h
index 0e18ad256f029b..27d82b20911cf9 100644
--- a/src/plugins/intel_gpu/src/graph/include/primitive_inst.h
+++ b/src/plugins/intel_gpu/src/graph/include/primitive_inst.h
@@ -67,8 +67,8 @@ struct primitive_impl {
     virtual std::vector<std::string> get_kernel_ids() {
         return {};
     }
-    virtual void save(cldnn::BinaryOutputBuffer& ob, const kernel_impl_params* impl_params = nullptr) const {}
-    virtual void load(cldnn::BinaryInputBuffer& ib, const kernel_impl_params* impl_params = nullptr) {}
+    virtual void save(cldnn::BinaryOutputBuffer& ob) const {}
+    virtual void load(cldnn::BinaryInputBuffer& ib) {}
 
     // If this flag is set as false, the memory allocated for this primitive is not allowed to be reused
     bool can_reuse_memory = true;
diff --git a/src/plugins/intel_gpu/src/graph/include/serialization/binary_buffer.hpp b/src/plugins/intel_gpu/src/graph/include/serialization/binary_buffer.hpp
index 7b19fb3ca2b142..1d211cc6d5caf4 100644
--- a/src/plugins/intel_gpu/src/graph/include/serialization/binary_buffer.hpp
+++ b/src/plugins/intel_gpu/src/graph/include/serialization/binary_buffer.hpp
@@ -24,8 +24,12 @@ class BinaryOutputBuffer : public OutputBuffer<BinaryOutputBuffer> {
         }
     }
 
+    void setKernlImplParams(void* impl_params) { _impl_params = impl_params; }
+    void* getKernlImplParams() const { return _impl_params; }
+
 private:
     std::ostream& stream;
+    void* _impl_params;
 };
 
 class BinaryInputBuffer : public InputBuffer<BinaryInputBuffer> {
@@ -39,8 +43,12 @@ class BinaryInputBuffer : public InputBuffer<BinaryInputBuffer> {
         }
     }
 
+    void setKernlImplParams(void* impl_params) { _impl_params = impl_params; }
+    void* getKernlImplParams() const { return _impl_params; }
+
 private:
     std::istream& stream;
+    void* _impl_params;
 };
 
 template <typename T>
diff --git a/src/plugins/intel_gpu/src/graph/primitive_inst.cpp b/src/plugins/intel_gpu/src/graph/primitive_inst.cpp
index 0859f777ec29ca..1a892f45b597f1 100644
--- a/src/plugins/intel_gpu/src/graph/primitive_inst.cpp
+++ b/src/plugins/intel_gpu/src/graph/primitive_inst.cpp
@@ -975,7 +975,8 @@ void primitive_inst::save(cldnn::BinaryOutputBuffer& ob) const {
         _impl->set_arguments(args_idx);
 
         _impl_params->save(ob);
-        _impl->save(ob, _impl_params.get());
+        ob.setKernlImplParams(_impl_params.get());
+        ob << _impl;
 
         ob << _node_output_layout;
         ob << has_mutable_input();
@@ -1100,7 +1101,8 @@ void primitive_inst::load(cldnn::BinaryInputBuffer& ib) {
         _impl_params = make_unique<kernel_impl_params>();
         _impl_params->load(ib);
         _impl.release();
-        _impl->load(ib, _impl_params.get());
+        ib.setKernlImplParams(_impl_params.get());
+        ib >> _impl;
 
         ib >> _node_output_layout;
         ib >> _has_mutable_input;