Interoperable GPU Kernels as Latency Improver for MEC