Was going to be a bit of a long comment so I've turned it into an answer. I am actually not sure of a good reference for this topic. Knapp's Lie Groups: Beyond an Introduction goes into real groups and covering groups a bit but I don't have a copy to check whether he discuss the metaplectic group or complexification in this manner.
The key here is that complexification is really a linear thing. We complexify real vector spaces and only from that can we obtain a way to complexify a group.
Throughout I will use the following definitions: a complexification of a real Lie group $G$ is a complex Lie group $G^\mathbb{C}$ such that $G \leq G^\mathbb{C}$ and the Lie algebra $\mathfrak{g}$ of $G$ is a real form of the Lie algebra $\mathfrak{g}^\mathbb{C}$. A universal complexification is a complex Lie group $G'$ equipped with a homomorphism $\phi: G \to G'$ with the universal property that any homomorphism from $G$ to a complex Lie group $H$, $f:G\to H$ there is a unique complex analytic homomorphism $F:G'\to H$ such that $f = F \circ \phi$. It should be clear that if a complexification exists then it is universal ($G^\mathbb{C} = G'$) with $\phi$ the inclusion map.
Firstly, the complexification. If $G \leq GL(n,\mathbb{R})$ then as you suggest it is not hard to find a complexification by simply restricting the exponential map $\exp:\mathfrak{gl}(n,\mathbb{C})\to GL(n,\mathbb{C})$ to $\mathfrak{g}^\mathbb{C}$ which will then extend the exponential map $\exp:\mathfrak{g}\to G$. In essence we are making use of the well conceived notion of complexification for $GL(n,\mathbb{R})$. Since you can define $\exp(X) = I + X + \frac{1}{2}X^2 + \dotsm$, you can even directly compute the exponential map on complex linear combinations of our real elements up to BCH formula computations etc. and certainly $\exp(iX) = I + iX - \frac{1}{2}X^2 + \dotsm$ makes sense.
It is crucial at this point to understand that a Lie algebra does not just have one exponential map. It has one for every Lie group it has and they are not the same thing. For example, $\exp:\mathfrak{sl}(n,\mathbb{R})\to SL(n,\mathbb{R})$ and
$\exp:\mathfrak{sl}(n,\mathbb{R})\to PSL(n,\mathbb{R})$ are different maps.
Secondly, the universal complexification.
As I discuss here you can construct $G'$ by first considering the homomorphism between the real and complex simply connected groups $\tilde{G}$ and $\tilde{G}_\mathbb{C}$, $\Phi:\tilde{G} \to \tilde{G}_\mathbb{C}$. Note the latter doesn't have to be the complexification of the former but we can still find such a $\Phi$ as the unique one whose differential is the inclusion map $\mathfrak{g}\hookrightarrow \mathfrak{g}^\mathbb{C}$. Simultaneously, we consider the projection $\pi: \tilde{G} \to G$ and define $G' := \tilde{G}_\mathbb{C}/N$ where $N$ is the smallest normal subgroup containing $\Phi(\ker \pi)$.
A few observations:
- This always produces a complex Lie group even when there is no complexification
- It can happen that the Lie algebra of $G'$ is not $\mathfrak{g}^\mathbb{C}$ but some quotient of it. This is not an issue in the semisimple case since $N$ will be discrete.
- If $G = \tilde{G}$ then $G' = \tilde{G}_\mathbb{C}$ since $\pi$ will be trivial.
As you can see this is a bit more of an abstract idea than the usual complexification but it always works.