Machine-to-machine communications has emerged to provide autonomic communications for a wide variety of intelligent services and applications. Among different communication technologies available for connecting machines, cellular-based systems have gained more attention as backhaul networks due to ubiquitous coverage and mobility support. The diverse ranges of service requirements as well as machine constraints require adopting different network architectures. This paper reviews three M2M network architectures to integrate machines into the LTE system and analyzes their associated communication delays. It also presents how the appropriate networks can be selected for some machine-to-machine applications, fulfilling their latency constraints.