I think the difficulty here will not be with setting up the FEM, but with the fact that this equation becomes singular on approaching the boundary x=0. [Dividing by x makes the coefficient of the drift term k/x.] One might get some useful insight by temporarily ignoring the variable y --- so looking first at the ODE: u''+(k/x)u'=f (which can be solved exactly) to see what behavior to expect as x->0.