The most common form of the self-consistent field (SCF) calculations, which are used in DFT and Hartree-Fock, assumes that each molecular orbital is a linear combination of all the atomic orbitals (aka basis function), and that the molecular orbitals are orthogonal to each other.
$$\psi^\mathrm{MO}_\mathrm{i} = \sum_\mathrm{p} c_\mathrm{pi} \chi^\mathrm{AO}_\mathrm{p}$$
So you start with $n$ atomic orbitals, and by taking linear combinations, you get $n$ orthogonal MO's. These molecular orbitals are called canonical orbitals. (Assuming there is no linear dependency in the AO basis set)
As the SCF procedure assumes from the start that each MO can have contributions from all of the AO's, the concept of delocalization is built into the procedure by default.
Therefore, when you do a DFT calculation on a molecule like benzene, no other considerations are required. If there is any delocalization, it will be treated by the method.
For example, have a look at the HOMO-1 orbital of butadiene (which is supposed to the lowest $\pi$-orbital, which all of the 2p-orbitals of carbon overlapping constructively) from a B3LYP/6-31+G(2d,p) calculation:

It is quite clear that the delocalization is accounted for.
As an aside, I want to add that even if you start with localized orbitals, the canonical wavefunction can be obtained by taking linear combination of the various different ways you can localize the orbitals. This is something I learnt in my quantum chemistry lectures, and the math works out, but it is difficult to have an intuitive sense of why this works. So for example you can write the structure of benzene in two Kekule forms, and it turns out that the canonical wavefunction of benzene is simply a linear combination of both - benzene exists as a superposition of the Kekule-type localized wavefunctions.
There can only be one set of canonical MO's that satisfy the orthonormality condition, but there can be multiple way to localize the MO's. There are actually some exotic quantum chemistry methods (such as ALMO), which actually start with localized wavefunctions (determined for specified fragments of the system), and then builds up the total wavefunction from those fragmented localized MO's. This works quite efficiently for large systems such as water clusters, because there is no delocalization, so electron-electron interaction only happens locally.