This PhD thesis proposes a new approach to classical Turing computability, called a model-theoretic approach. In that approach, structures and theories are associated to Turing machines in order to study the characteristics of their computations. A model-theoretic approach to Turing computability through first-order logic is developed, and first results about correspondence, soundness, representation and completeness among Turing machines, structures and theories are proved. In this line, the results about properties as stability, absoluteness, universality and logicality emphasize the importance of the model-theoretic standpoint. It is shown that the underlying logic of Turing theories is a minimal intuicionistic logic, being able to internalize a classical negation operator. The techniques obtained in the present dissertation permit us to examine the Turing computability over nonstandard models of arithmetic as well. In this context, a new perspective about Tennenbaum's phenomenon and a critical evaluation of Dershowitz and Gurevich's account on Church-Turing's thesis are given. As a consequence, an arithmetic internality principle is postulated, according to which the concept of computation itself is relative to the arithmetic model that Turing machines operate. In this way, the dissertation unifies the existing model-arithmetic characterizations of the P versus NP problem, leading, as a by-product, to a model-arithmetic barrier to the solvability of that central problem in computational complexity with respect to certain techniques. As a whole, the dissertation sustains that crucial characteristics of the concept of computation may be understood from the duality between finiteness and infiniteness inherent within the distinction between standard and nonstandard natural numbers.