Text this: Enhancing DNN Computational Efficiency via Decomposition and Approximation