With the increase in urban complexity, plausible analytical and synthetic models became highly valued as the way to decode and reconstruct the organization that makes urban systems. What they lacked is a mechanism by which an analytical description of urban complexity could be translated into a synthetic description. An attempt to define such a mechanism is presented in this paper, where knowledge is retrieved from the natural organization that cities settle into, and devised in a design model to support urban design at the problem definition stage. The model comprises two automated design modules, giving preference to street accessibility. The first module implements plausible spatial laws to generate street structures. The performance criteria of these structures are measured against accessibility scores and clustering patterns of street segments. In the second module, an Artificial Neural Networks model (ANNs) is trained on Barcelona's data, outlining how street width, building height, block density and retail land use might be dependent on street accessibility. The ANNs is tested on Manhattan's data. The application of the two computational modules is explored at the problem definition stage of a design process in order to verify how far deterministic knowledge-based models are in the transition from the analysis of design problems to the synthesis of design solutions. Our findings suggest that the computational framework proposed could be instrumental at generating simplified representation of an urban grid, whilst being effective at forecasting form-related and functional attributes within a minimum resolution of 200 meters. It is finally concluded that as design progresses, knowledge-based models may serve as to minimize uncertainty about complex urban design problems.