diff --git a/opencl/source/mem_obj/mem_obj.cpp b/opencl/source/mem_obj/mem_obj.cpp
index e31c75d2a8..0665cbf0d1 100644
--- a/opencl/source/mem_obj/mem_obj.cpp
+++ b/opencl/source/mem_obj/mem_obj.cpp
@@ -94,6 +94,8 @@ MemObj::~MemObj() {
                 }
                 destroyGraphicsAllocation(graphicsAllocation, doAsyncDestructions);
                 graphicsAllocation = nullptr;
+            } else if (graphicsAllocation && context->getBufferPoolAllocator().isPoolBuffer(associatedMemObject)) {
+                memoryManager->waitForEnginesCompletion(*graphicsAllocation);
             }
             if (!associatedMemObject) {
                 releaseMapAllocation(rootDeviceIndex, doAsyncDestructions);
diff --git a/opencl/test/unit_test/mem_obj/buffer_pool_alloc_tests.cpp b/opencl/test/unit_test/mem_obj/buffer_pool_alloc_tests.cpp
index 0e09610a28..0ebed29e04 100644
--- a/opencl/test/unit_test/mem_obj/buffer_pool_alloc_tests.cpp
+++ b/opencl/test/unit_test/mem_obj/buffer_pool_alloc_tests.cpp
@@ -115,6 +115,25 @@ TEST_F(aggregatedSmallBuffersEnabledTest, givenAggregatedSmallBuffersEnabledAndS
     EXPECT_EQ(retVal, CL_SUCCESS);
 }
 
+TEST_F(aggregatedSmallBuffersEnabledTest, givenAggregatedSmallBuffersEnabledWhenClReleaseMemObjectCalledThenWaitForEnginesCompletionCalled) {
+    ASSERT_TRUE(poolAllocator->isAggregatedSmallBuffersEnabled());
+    ASSERT_NE(poolAllocator->mainStorage, nullptr);
+    std::unique_ptr<Buffer> buffer(Buffer::create(context.get(), flags, size, hostPtr, retVal));
+
+    ASSERT_NE(buffer, nullptr);
+    ASSERT_EQ(retVal, CL_SUCCESS);
+
+    ASSERT_NE(poolAllocator->mainStorage, nullptr);
+    auto mockBuffer = static_cast<MockBuffer *>(buffer.get());
+    ASSERT_TRUE(mockBuffer->isSubBuffer());
+    ASSERT_EQ(poolAllocator->mainStorage, mockBuffer->associatedMemObject);
+
+    ASSERT_EQ(mockMemoryManager->waitForEnginesCompletionCalled, 0u);
+    retVal = clReleaseMemObject(buffer.release());
+    ASSERT_EQ(retVal, CL_SUCCESS);
+    EXPECT_EQ(mockMemoryManager->waitForEnginesCompletionCalled, 1u);
+}
+
 TEST_F(aggregatedSmallBuffersEnabledTest, givenCopyHostPointerWhenCreatingBufferButCopyFailedThenDoNotUsePool) {
     class MockCommandQueueFailFirstEnqueueWrite : public MockCommandQueue {
       public: