Although random linear network coding (RLNC) constitutes a highly efficient and distributed approach to enhance communication networks and distributed storage, it requires additional processing to be carried out in the network and in end devices. For mobile devices, this processing translates into energy use that may reduce the battery life of a device. This paper focuses not only on providing a comprehensive measurement study of the energy cost of RLNC in eight different computing platforms, but also explores novel approaches (e.g., tunable sparse network coding) and hardware optimizations for Single Instruction Multiple Data (SIMD) available in the latest generations of Intel and Advanced RISC Machines (ARM) processors. Our measurement results show that the former provides gains of two-To six-fold from the underlying algorithms over RLNC, while the latter provides gains for all schemes from 2x to as high as 20x. Finally, our results show that the latest generation of mobile processors reduce dramatically the energy per bit consumed for carrying out network coding operations compared to previous generations, thus making network coding a viable technology for the upcoming 5G communication systems, even without dedicated hardware.