With autonomous driving, vehicles are undergoing tremendous and multiple innovations in a variety of areas of automotive expertise. In particular, the amount of software used in embedded safety-critical systems is increasing at a rapid rate to implement new features. It is therefore essential today to guarantee the safety of software by carrying out safety analyses in accordance with automotive standards. These analyses allow engineers assessing the design with regard to safety and to determine the modifications if needed to meet safety objectives. However, the traditional approach to perform these analyses is cumbersome and limited when faced with the complexity of today's automotive software architectures. Safety analyses are currently performed manually, and the results are dependent on the experience of the safety expert. As a result, they are highly subjective and are not guaranteed to be exhaustive and error-free. To overcome these issues, this paper explores the use of a model-based safety approach in the context of safety-critical automotive embedded software. It makes a methodological proposal that relies on the software architecture model to build a dedicated safety model from which safety analyses can be automatically derived. The method is experimented on an automotive case study, an embedded software that assists the driver in following the lane.