The major difference between sql server 2012 and below vs sql server 2014 and up is the new cardinality estimator (CE)
See my answer here as well.
My Story
When we migrated from sql server 2008R2 to 2014, we saw huge query regressions happening on our application. So we decided to enable TF 9481 but still have the latest compatiblity level.
Upgrading from sql server 2014 to 2016, I did not see any query regressions. So no Traceflag or older compatiblity mode required in sql server 2016.
In sql server 2016, there is database scope configuration option to govern the CE and optimizer fixes (equivalent of TF 4199)

Should I expect differences in performance after the
compatibility level change?
Its a big change for your application. SO test it properly. You can use Database Experimentation Assistant (New & in preview)
What is the best way to measure the performance before and after?
You can use Query store (new in sql server 2016) to measure performance and find easily regressed queries interms of IO, CPU, etc.
Should I take special care when moving the databases the SQL
server 2016?
I have written lenghty answer for migrating from lower version of sql server to higher version of sql server.