Without initial or boundary conditions, the solution to your average DE is a family of solutions. For example, the solution set to $x' = x$ is $\{Ae^t\}$, a set parametrised by the value $A$.
So, if $x_p$ is a particular solution to a given linear DE, and $x_h$ is a generic element of the solution set of the homogeneous DE, then at the very least you can show that $x = x_h + x_p$ represents a solution set to the original DE, because $x' = x_h' + x_p'$ and so on for all further derivatives. In other words, if the original DE is $a_0 x + a_1 x' + a_2 x'' + \ldots + a_n x^{(n)} = c$, then we have:
$\begin{eqnarray}a_0 x + a_1 x' + a_2 x'' + \ldots + a_n x^{(n)} & = & a_0 (x_h + x_p) + a_1 (x_h + x_p)' + \ldots + a_n (x_h + x_p)^{(n)} \\ & = & a_0 x_h + a_0 x_p + a_1 x_h' + a_1 x_p' + \ldots + a_n x_h^{(n)} + a_n x_p^{(n)} \\ & = & \left( a_0 x_h + a_1 x_h' + \ldots + a_n x_h^{(n)} \right) + \left( a_0 x_p + a_1 x_p' + \ldots + a_n x_p^{(n)} \right) \\ & = & 0 + C\end{eqnarray}$
Because $x_h$ is a solution to the homogeneous DE $a_0 x_h + a_1 x_h' + \ldots + a_n x_h^{(n)} = 0$ and $x_p$ is a solution to the linear DE.
Thus, we at least know that if we add a particular solution and a homogeneous solution together, we get another solution to the DE. So if we add the entire homogeneous solution set to the particular solution we generate a set of functions that are also solutions to the original DE. So as long as we can then pick an element of that set that satisfies our boundary/initial conditions, we can generate the required solution to the DE.
Proof that (a) the set of all $x_h + x_p$ is the full solution set to the DE, and (b) that the method generates unique solutions are slightly trickier exercises.