A method for determining an order in which to stream modules of an application from a server to a client, where the application uses modules in an execution-time dependent order, employs a predictive model to determine which module transitions are most likely to occur in the future in view of the clients present state in the application and then stream the modules to the client in the determined order. Different sets of transition weights can be used for clients falling into different categories, such as time of application use and age of the user at the client. Additional factors which can be used to select the order to stream the modules include cache utilization at the client and customized streaming orders specified by the user.