diff --git a/inference-engine/thirdparty/clDNN/kernel_selector/core/cl_kernels/pooling_gpu_b_fs_zyx_fsv16_imad.cl b/inference-engine/thirdparty/clDNN/kernel_selector/core/cl_kernels/pooling_gpu_b_fs_zyx_fsv16_imad.cl
index 3b55a1a401c72a..ca8342d122507a 100644
--- a/inference-engine/thirdparty/clDNN/kernel_selector/core/cl_kernels/pooling_gpu_b_fs_zyx_fsv16_imad.cl
+++ b/inference-engine/thirdparty/clDNN/kernel_selector/core/cl_kernels/pooling_gpu_b_fs_zyx_fsv16_imad.cl
@@ -126,26 +126,45 @@ KERNEL(pooling_gpu_b_fs_zyx_fsv16)(
         ACTIVATION_VEC16 pool_result;
 #if defined AVG_POOLING
 #if INPUT0_FEATURE_NUM % FEATURE_SLICE_SIZE != 0
-        __attribute__((opencl_unroll_hint(INPUT0_FEATURE_NUM % FEATURE_SLICE_SIZE)))
-        for (uint i = 0; i < INPUT0_FEATURE_NUM % FEATURE_SLICE_SIZE; i++) {
+        if (last_in_f_group) {
+            __attribute__((opencl_unroll_hint(INPUT0_FEATURE_NUM % FEATURE_SLICE_SIZE)))
+            for (uint i = 0; i < INPUT0_FEATURE_NUM % FEATURE_SLICE_SIZE; i++) {
+#if ENABLE_ROUND
+#if defined(DYNAMIC_KERNEL_DIVIDER) || defined(DYNAMIC_WITH_PADDING_KERNEL_DIVIDER)
+                pool_result[i] = convert_int(round(((float)result[i] / max(num_elements, (uint)1))));
 #else
-        __attribute__((opencl_unroll_hint(FEATURE_SLICE_SIZE)))
-        for (uint i = 0; i < FEATURE_SLICE_SIZE; i++) {
+                pool_result[i] = convert_int(round((float)result[i] / (int)(POOL_SIZE_X * INPUT0_SIZE_Z * INPUT0_SIZE_Y)));
+#endif
+#else // ENABLE_ROUND
+#if defined(DYNAMIC_KERNEL_DIVIDER) || defined(DYNAMIC_WITH_PADDING_KERNEL_DIVIDER)
+                pool_result[i] = (float)result[i] / max(num_elements, (uint)1);
+#else
+                pool_result[i] = (float)result[i] / (int)(POOL_SIZE_X * INPUT0_SIZE_Z * INPUT0_SIZE_Y);
 #endif
+#endif  // ENABLE_ROUND
+            }
+        } else {
+#endif
+            __attribute__((opencl_unroll_hint(FEATURE_SLICE_SIZE)))
+            for (uint i = 0; i < FEATURE_SLICE_SIZE; i++) {
+// INPUT0_FEATURE_NUM % FEATURE_SLICE_SIZE != 0
 #if ENABLE_ROUND
 #if defined(DYNAMIC_KERNEL_DIVIDER) || defined(DYNAMIC_WITH_PADDING_KERNEL_DIVIDER)
-            pool_result[i] = convert_int(round(((float)result[i] / max(num_elements, (uint)1))));
+                pool_result[i] = convert_int(round(((float)result[i] / max(num_elements, (uint)1))));
 #else
-            pool_result[i] = convert_int(round((float)result[i] / (int)(POOL_SIZE_X * INPUT0_SIZE_Z * INPUT0_SIZE_Y)));
+                pool_result[i] = convert_int(round((float)result[i] / (int)(POOL_SIZE_X * INPUT0_SIZE_Z * INPUT0_SIZE_Y)));
 #endif
 #else // ENABLE_ROUND
 #if defined(DYNAMIC_KERNEL_DIVIDER) || defined(DYNAMIC_WITH_PADDING_KERNEL_DIVIDER)
-            pool_result[i] = (float)result[i] / max(num_elements, (uint)1);
+                pool_result[i] = (float)result[i] / max(num_elements, (uint)1);
 #else
-            pool_result[i] = (float)result[i] / (int)(POOL_SIZE_X * INPUT0_SIZE_Z * INPUT0_SIZE_Y);
+                pool_result[i] = (float)result[i] / (int)(POOL_SIZE_X * INPUT0_SIZE_Z * INPUT0_SIZE_Y);
 #endif
 #endif  // ENABLE_ROUND
+            }
+#if INPUT0_FEATURE_NUM % FEATURE_SLICE_SIZE != 0
         }
+#endif // INPUT0_FEATURE_NUM % FEATURE_SLICE_SIZE != 0
 #else   // AVG_POOLING
         pool_result = TO_ACTIVATION_VEC16(result);
 #endif  // AVG_POOLING
diff --git a/inference-engine/thirdparty/clDNN/tests/test_cases/pooling_gpu_test.cpp b/inference-engine/thirdparty/clDNN/tests/test_cases/pooling_gpu_test.cpp
index 8650127f4085ed..ac5629a8f01c3d 100644
--- a/inference-engine/thirdparty/clDNN/tests/test_cases/pooling_gpu_test.cpp
+++ b/inference-engine/thirdparty/clDNN/tests/test_cases/pooling_gpu_test.cpp
@@ -2719,7 +2719,7 @@ INSTANTIATE_TEST_SUITE_P(
     smoke_low_precision_2d_spatial,
     pooling_random_test,
     testing::Combine(testing::Values(1, 2),
-                     testing::Values(3, 8, 64),
+                     testing::Values(3, 8, 19, 64),
                      testing::Values(std::tuple<size_t, size_t, size_t>(12, 12, 1), std::tuple<size_t, size_t, size_t>(24, 24, 1)),
                      testing::Values(std::tuple<size_t, size_t, size_t>(4, 4, 1), std::tuple<size_t, size_t, size_t>(2, 2, 1)),
                      testing::Values(std::tuple<int, int, int>(2, 2, 1)),
@@ -2736,7 +2736,7 @@ INSTANTIATE_TEST_SUITE_P(
     smoke_low_precision_3d_spatial,
     pooling_random_test,
     testing::Combine(testing::Values(1, 2),
-                     testing::Values(3, 8, 64),
+                     testing::Values(3, 8, 27, 64),
                      testing::Values(std::tuple<size_t, size_t, size_t>(12, 12, 12), std::tuple<size_t, size_t, size_t>(24, 24, 24)),
                      testing::Values(std::tuple<size_t, size_t, size_t>(4, 4, 4), std::tuple<size_t, size_t, size_t>(2, 2, 2)),
                      testing::Values(std::tuple<int, int, int>(2, 2, 2)),