Preferential crystallization is a rather cheap alternative to separate mixtures of enantiomers into the pure chiral species. Especially in the pharmaceutical industry the separation of enantiomers is of growing importance. Despite its rather low costs compared with other separation methods (e.g. chromatography, membrane processes) the application of the method in the industrial practice is scarce. One reason for this could be the assumed liability to disturbances of the process due to its kinetically controlled nature. In this context the thesis provides a contribution to the understanding of the process as well as a guideline to process design. One goal of the work is to increase the usage of the preferential crystallization process as an alternative to other separation methods. Apart from that the methods introduced can also be applied to other (crystallization) processes. At first a concept for an a priori process evaluation based on solubilities and metastable zone width is introduced. The process concept is then tested and evaluated for two different systems. Using the example of the conglomerate forming system DL-threonine/water a more complex dynamic modeling approach based on population balances is introduced. Based on measured separation runs the free kinetic parameters of the used models are estimated. The reliability of the parameter estimates is evaluated using the Fisher information matrix or a Bootstrap method respectively. Additional experiments are designed based on the developed models using a dynamic experimental design. Using this approach the number of experimental runs can be minimized. As a prerequisite for process modeling and optimization an online and inline analytic is established and calibrated which is used to investigate the liquid phase composition as well as properties of the particle size distribution. Depending on the available information and the process design goal a reduced moment model or a fully discretized model is used. The reduced model offers the advantage of increased computational speed, whereas the fully discretized model does not only provide information with respect to the moments of the particle size distribution but provides the full particle size distribution at every time point. The developed kinetic model is implemented into Matlab® as well as into Parsival®. The different model solution strategies are compared and therefore different options for model simulation are provided. Finally the parameterized and validated model implemented into Parsival® is used to optimize the process in terms of productivity. Additionally the model can be used to predict the mean diameter of the product crystals and the variance of the particle size distribution for different experimental conditions.