I think that any of these functions, Gaussian, Lorenzian, Voight, will do, they do not differ by much but the Lorenzian has a slightly wider base and in overlapping spectra this may increase background too much. I assume that you will convolute your stick spectrum to get your 'real' spectrum, in this case you can zero pad the transform, it makes nicer looking spectra but adds no more information.
Perhaps the most important point to check is the spectral resolution of the instrument you are using, and adjust the width of you broadening function appropriately.

I think that any of these functions, Gaussian, Lorenzian, Voight, will do, they do not differ by much but the Lorenzian has a slightly wider base and in overlapping spectra this may increase background too much. I assume that you will convolute your stick spectrum to get your 'real' spectrum, in this case you can zero pad the transform, it makes nicer looking spectra but adds no more information.
Perhaps the most important point to check is the spectral resolution of the instrument you are simulating spectra from, and adjust the width of you broadening function appropriately.