Cellular learning automata (CLAs) have received significant attentions by researchers in recent years. CLAs are learning models, which bring together the computational power of cellular automata (CAs), and the learning capability of learning automata (LAs). From the viewpoint of the structure regularity, CLAs can be classified into two classes: irregular CLAs and regular CLAs. In an irregular CLA, the structure regularity assumption is relaxed. In contrast to regular CLAs, irregular CLAs have irregular structure, which is more suitable for problem solving in some areas such as ad hoc and sensor networks, internet of things, and grid computing. As another classification, from the viewpoint of structure dynamicity, CLAs can be also classified into static CLAs and dynamic CLAs. In a static CLA, the underlying graph of the cellular structure remains fixed during the evolution of the CLA. In a dynamic CLA, one of its aspects such as structure, local rule, or neighborhood set may change over time. In comparison to static CLAs, dynamic models of CLAs are more suitable for modeling and problem solving of domains that are dynamic in nature, such as peer-to-peer, internet of things and mobile sensor networks. In this chapter, irregular CLAs as well as dynamic models of CLAs will be proposed. Furthermore, the steady-state behavior of the proposed models will be studied, notion of expediency will be defined for them and conditions under which these models become expedient will be established.