In order to quickly implement an embedded system that is mainly based on software, two orthogonal approaches have been proposed: Platform-based Design, which maximizes the reuse of components; and Model Driven Development, which rises the abstraction level by using object-oriented concepts and UML. However, with this increasing of the abstraction level, software engineers do not have an exact idea of the impact of their modeling decisions on important issues such as performance, energy, and memory footprint for a given platform. In our approach, analytical estimation of data and program memory, performance, and energy are obtained directly from UML models. Experimental results show a very small estimation error when software components are reused and their costs on the target platform are already known. Real-life applications are modeled in different ways and demonstrate the effectiveness of the estimates in an early design space exploration, allowing the designer to quickly compare different modeling solutions, with estimation errors as low as 5%.