Water-soluble crystals with anti-caking function have become high-end products, whereas they are prepared by trial and error due to the lack of quantitative correlation between particle size and caking tendency, as well as the absence of the models for global consideration between caking and crystallization. This work presents a coupled Caking-Size-Crystallization (CSC) model, which connects the front-end crystallization process and the post-end caking process to achieve the closed-loop design of the anti-caking function for the first time. The CSC model can directly generate the optimal cooling trajectory to crystallize the product free of caking. The introduction of the caking constant term, the mean particle size and the Mullin-Nyvlt’s theory improve the design efficiency by an order of magnitude with the precision guaranteed. This model-based approach presents an opportunity to find optimal crystallization operating profiles with diverse crystal seed loading strategies that meet both product function and manufacturability demands.