I certainly wouldn't want to see corporate devs at my bank implement some custom "secure" http communications protocol.
What we need is two or three well funded TLS/SSL implementations competing with each other, kind of like we have a handful of competing web browser implementations or C++ compiler implementations or operating systems, all done by experts in their field.
I don't know PolarSSL. If it's a reasonably complete implementation then maybe it deserves more attention.
The problem with Telegram wasn't about code, it was about algorithms and design. They designed their own crypto, which is a really bad thing. If they had written their own implementations of well-known, tested and proven crypto primitives/algorithms, everything would have been fine (well, mostly; you can still implement a good algorithm badly, but that's not what happened).