Novel methodologies for predictable CPU-to-GPU command offloading