Nowadays, 3D multimedia applications have grown rapidly in number and consist of complex systems (e.g. 3D graphical processing or games) that process extensive amounts of data to create 3D images and results. This produces highcost and high-power consumption systems whereas a superior portability demands cheap and low-power consumption ones. In these multimedia applications, the dynamic memory subsystem is currently one of the main sources of power consumption and its inattentive management can affect severely the performance and power consumption of the whole system. In this paper, we illustrate a new system-level method to explore and refine the dynamic memory management of multimedia systems on current typical case studies, i.e. a relatively new 3D image reconstruction system and a 3D simulation game. This method is based on an analysis of the access pattern, amount of memory used and power consumption estimations. With this information, a phasewise exploration and refinement flow is used to optimize the system at the different phases of its hardware-oriented design process. As the results in the case studies show, our system-level method achieves great improvements in memory footprint, power consumption and performance for multimedia applications.