Photonic nanostructures have attracted a tremendous amount of attention in the recent past. Via their size, shape and material it is possible to engineer their optical response to user-defined needs. Tailoring of the optical response is usually based on a reference geometry for which subsequent variations to the initial design are applied. Such approach, however, might fail if optimum nanostructures for complex optical responses are searched. As example we can mention the case of complex structures with several simultaneous optical resonances. We propose an approach to tackle the problem in the inverse way: In a first step we define the desired optical response as function of the nanostructure geometry. This response is numerically evaluated using the Green Dyadic Method for fully retarded electro-dynamical simulations. Eventually, we optimize multiple of such objective functions concurrently, using an evolutionary multi-objective optimization algorithm, which is coupled to the electro-dynamical simulations code. A great advantage of this optimization technique is, that it allows the implicit and automatic consideration of technological limitations like the electron beam lithography resolution. Explicitly, we optimize silicon nanostructures such that they provide two user-defined resonance wavelengths, which can be individually addressed by crossed incident polarizations.