We switched to Decoder due to allocations after profiling - we were using the Decoders/Encoders in a pipeline.

Don't believe there was a more convenient way to achieve the same without writing our own Decoders/Encoders.