High resource consumption and latency during deep neural network execution prevent deployment on constrained hardware. Reducing these bottlenecks allows for real-time processing and efficient static execution planning.