As modern vehicle architectures grow increasingly complex, the automotive industry relies on Model-Based Systems Engineering (MBSE) to design advanced systems and ensure traceability. MBSE adopts a structured approach, integrating requirements with functional, logical, and physical views to support cohesive system design. This study investigates methods to automate the generation of initial functional architectures from requirements using advanced clustering techniques. Although clustering has been extensively applied to non-functional requirements (NFRs), its application to functional requirements (FRs) in automotive contexts remains underexplored. This research addresses this gap by transforming functional requirements into dense vectors through modern embedding models, such as Bidirectional Encoder Representations from Transformers (BERT), Embeddings from Language Models (ELMo), Global Vectors for Word Representation (GloVe), and Doc2Vec, to capture their complex semantic relationships. These transformations enable more precise clustering through Agglomerative Hierarchical Clustering (AHC). The results demonstrate that BERT outperformed other embedding methods, achieving superior clustering accuracy and precision. These findings emphasize the potential of large language models (LLMs) to advance the clustering and classification of FRs by enhancing their contextual understanding and scalability. Overall, this study demonstrates that leveraging advanced Natural Language Processing (NLP) techniques enhances the management of FRs in automotive systems, enabling more effective decision-making and improved design efficiency.