The admissible representations of U(2) over a p-adic field are fairly straightforward since they essentially come from those of SL(2) or a the norm one elements of the quaternions.
Let me add (two days later) that the point is not that the representation of U(3) that arises from the theta-lift is non-tempered but rather that is non-tempered AND the way it arises from a theta-lift. This allows you to construct CAP representations of U(3) globally.