It seems that my newsreader doesn't know how to reply to cross-posted messages, so here is my reply from the EMF newsgroup.
-----8<-----
Hi, Gilbert,
There is no AssociationEnd metaclass: it was removed in UML 2.0. Instead, association ends are represented by the Property metaclass. So, your Event stereotype needs to extend Property.
To assert that the ends of a <<dataChannel>> association are <<events>> you will need to use OCL. e.g.: