In the context of mobile handheld devices, energy consumption is a primary concern and the process of video decoding is often among the most resource-intensive applications. Recent embedded processors are equipped with advanced features such as Dynamic Voltage Frequency Scaling (DVFS) in order to reduce their power consumption. These features can be used to perform low power video decoding when no hardware decoding support is available for a given standard. High Efficiency Video Coding (HEVC) is a recent video standard offering state-of-the-art compression rates and advanced parallel processing solutions. This paper presents strategies for the power optimization of a real-time software HEVC decoder on Neon architecture. These strategies include the exploitation of data and task-level parallelism, as well as the use of a new frequency control system to optimize the processor DVFS, based on an estimation of the decoding complexity. Extensive power measurement results, based on a multi-core ARM big.LITTLE processor, are provided and compared to state-of-the-art. These results show that the proposed open-source implementation can reach an energy consumption below 21 nJ/px for HD decoding at 2.2 Mbits/s.