Accelerating Deep Learning Inference via Model Parallelism and Partial Computation Offloading | Synapse