TLS is a pile of IETF design-by-committee extensions uncomfortable with saying "no;" OpenSSL is a mountainous kitchen sink of TLS and other features. OpenSSL seems to "work," but the endless pattern of vulnerabities lies in the risky process in which code is developed, lightly tested and pushed out as a production release... That hasn't changed enough to convince me they're going to get proactive and stop SNAFUBAR anytime soon.
'cperciva was on the right track about a better protocol would encode parameters much more simply and have just enough options for an important gamut of use-cases. The issue is we're seemingly stuck on TLS everywhere and would require massive cost and coordination pains to use something, even if it were 10X better and formally-proven.