First thing to look at is:

Is "latch free" a dominant part of your overall response time. If my application takes 5 seconds to respond, and 0.01 seconds of that is "latch free", then why waste time looking there.

Lets assume it *is* a dominant part. If you are waiting for latch free, the someone else is *holding* that a latch that you want. So the first two things to do:

1) what latch are you waiting for ?
2) who is holding it

For (1), check your AWR reports. There's a section which shows each latch, and how many gets/misses/sleeps are occurring. The most common ones I see are "shared pool" and "library cache", which is typically indicative of excessive parsing in queries.

For (2), there is a view v$latchholder. You can sample this with Tanel's excellent script

https://blog.tanelpoder.com/files/scripts/latchprof.sql

to deduce *who* is taking those latches so aggressively.