My Question is now how the RTP Media Stream should/can flow. The clients are in different other networks. So P2P Media Stream isn't possible. Should I now run the RTP Stream Client - Asterisk or Client - Kamailio - Asterisk?
What do you want to accomplish?
Fact is that asterisk has to handle RTP in case of transcoding or DTMF in RTP, reinvite/directmedia doesn't work in all scenarios. So why spend extra time/resources on relaying it on the kamailio machine with rtp(proxy|engine)? Having said this, it is much easier to maintain an asterisk cluster behind a machine with rtp(proxy|engine) doing NAT traversal.