Some new error-resilient source coding and joint source/channel coding techniques are proposed for the transmission of multimedia sources over error-prone channels. First, we introduce a class of entropy codes providing unequal error-resilience, i.e. providing some protection to the most sensitive information. These codes are then extended to exploit the temporal dependencies. A new state model based on the aggregation of some states of the trellis is then proposed and analyzed for soft source decoding of variable length codes with a length constraint. It allows the weighting of the compromise between the estimation accuracy and the decoding complexity. Next, some paquetization methods are proposed to reduce the error propagation phenomenon of variable length codes. Finally, some re-writing rules are proposed to extend the binary codetree representation of entropy codes. The proposed representation allows in particular the design of codes with improved soft decoding performances.