Support for matrix exp

```julia
using CuArrays
n = 4
A = randn(n, n) |> cu
exp(A)
```
```
ERROR: ReadOnlyMemoryError()
Stacktrace:
[1] exp!(::CuArray{Float32,2}) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v0.7/LinearAlgebra/src/lapack.jl:209
[2] exp(::CuArray{Float32,2}) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v0.7/LinearAlgebra/src/dense.jl:508
[3] top-level scope at none:0
```

I get the same error without a stack trace for `A \ cu(randn(n))`

(This is using CuArrays#vc/0.7 (a55b12a180273c80c34c1d) depending on GPUArrays#master, since CuArrays#master doesn't seem to support indexing `A[i,j]` that `exp(m::AbstractMatrix)` uses)

How much work would it take to support LinearAlgebra functions like these? If reasonable for a beginner like myself, any pointers to where to start?

Thanks!