Next generation virtual radio access networks (VRAN) will benefit from the flexibility provided by virtualization in proposed CloudRAN configurations. These systems for 5G and beyond may consist of commodity hardware such as GPUs in data centers with multiple connected basestations (gNBs) flexibly receiving allocated resources depending on time-varying real-time demands. In this paper, parallel reconfigurable algorithms and architectures for channel decoding are proposed. In particular, flexible rate and block length LDPC decoders for the new radio (NR) physical layer on GPU are characterized. We implement these GPU decoders using reduced word lengths of 8-bits to represent the log-likelihood ratios during decoding, and we utilize multiple GPU streams to process multiple blocks of codewords in parallel. These techniques allow our implementation to reduce the device transfer overhead and achieve low-latency targets for 5G and beyond. Moreover, we integrate our decoder into the Open Air Interface (OAI) NR software stack to investigate virtualization capabilities when containerizing VRAN functionality such as the LDPC decoder.