I think this will work, but you'd need to have named instances. You can't have things fail from 2016-> 2014. This should be like any other cluster that has multiple instances, potentially limiting which services can reside on which nodes.
Can I ask why a single cluster? Are you planning an upgrade? Why not two?