NAT differentiation between TCP and UDP.

I'm trying to implement a TCP hole punching example, and I need to know whether NAT devices use the same bindings for UDP and TCP sessions or keep different records for each. Whether such behaviour is specified by a standard or is implementation-dependent. I've skimmed the RFC, but it's unclear on this point.

For example, would a TCP packet from say 192.168.10.10:12345 use the same binding as a previous UDP session from 192.168.10.10:12345 or cause a new binding to be created. As far as I can tell there would be no problem using the same entry, both bindings lead back to same host and it can send the UDP packets to the UDP endpoint, and the TCP packets to the TCP endpoint easily enough.