I have done the geometry optimization with the B3LYP functional, however, there is a need to calculate the excited state energy, so the CAM-B3LYP functional should be better suited for that. Do I need to reoptimize the geometry to use CAM-B3LYP or are these functionals simillar enough so no reoptimization is needed?
What is sometimes done is use the geometry and thermochemical corrections (i.e. the results of the frequency calculation) of method X, but use the electronic energy (SCF energy) of another method Y. This is generally noted
Y//X. Since the geometries and thermochemical corrections typically are not much influenced by the quality of the method (beyond a certain point), this scheme gives good result. The electronic energy can however benefit from a larger basis set or a better computational method (e.g. MP2 with DFT geometries).
I am assuming this is what you have seen in the literature, as it is relatively common, especially for large molecules like organometallic complexes. I am not too familiar with excited states, but the same thing likely applies.
However, you must always use the exact same method for the geometry optimisation and the frequency calculation. The frequency calculation uses the curvature of the potential energy surface (PES) to calculate approximate corrections. If you are not at an extremum (minimum for equilibrium structures or saddle point for transition state), the results are meaningless. Changing the method will move you off the extremum sufficiently and your frequency calculation will likely have several imaginary frequencies.