That's a lot harder. You literally just need two models or simply render the head part manually with the applied texture. And then you just need to create the rendertype with the texture twice. It's literally just what the LivingEntityRenderer is doing twice.
The pixel information is stored in BGRA format, so that makes sense.