Yeah, And it would be very cool if the kernel could do this automatically, for missing instructions if required.

I have stumbled upon a new problem nowadays:
Let's say that you have a specific CPU from 2012 that lacks the specific AVX instruction set that a new program requires to run

format PE64 NX GUI 6.0
entry start

include 'win64ax.inc'

section '.data' data readable writeable
    ; [!] IMPORTANT: Always use ABSOLUTE PATHS for injection. 
    ; The target process might have a different Current Working Directory.
    target_path db 'C:\Path\To\TargetApp.exe', 0
    dll_path    db 'C:\Path\To\YourVEH.dll', 0 
    dll_len     = $ - dll_path

    si          STARTUPINFO
    pi          PROCESS_INFORMATION
    
    remote_mem  dq ?     ; Address of string in target process
    h_inj_th    dq ?     ; Handle to injection thread
    
    k32_lib     db 'kernel32.dll', 0
    loadlib_fn  db 'LoadLibraryA', 0

section '.code' code readable executable

start:
    ; Initialize startup info size (Critical for CreateProcess)
    mov     [si.cb], sizeof.STARTUPINFO

    ; 1. Launch Target in SUSPENDED state
    ;    The main thread halts at RtlUserThreadStart (before any app code runs).
    invoke  CreateProcessA, 0, target_path, 0, 0, FALSE, \
            CREATE_SUSPENDED, 0, 0, si, pi
    
    test    rax, rax
    jz      .exit

    ; 2. Allocate memory in target for the DLL path
    invoke  VirtualAllocEx, [pi.hProcess], 0, dll_len, \
            MEM_COMMIT or MEM_RESERVE, PAGE_READWRITE
    
    mov     [remote_mem], rax
    test    rax, rax
    jz      .cleanup

    ; 3. Write the DLL path string into the target
    invoke  WriteProcessMemory, [pi.hProcess], [remote_mem], \
            dll_path, dll_len, 0

    ; 4. Resolve LoadLibraryA address
    ;    (Safe assumption: Kernel32 is mapped at the same address in all processes)
    invoke  GetModuleHandleA, k32_lib
    invoke  GetProcAddress, rax, loadlib_fn
    
    ; 5. Create Remote Thread to execute LoadLibraryA("YourVEH.dll")
    invoke  CreateRemoteThread, [pi.hProcess], 0, 0, \
            rax,            \ ; lpStartAddress (LoadLibraryA)
            [remote_mem],   \ ; lpParameter (Path String)
            0, 0
            
    mov     [h_inj_th], rax
    test    rax, rax
    jz      .cleanup

    ; 6. Wait for Injection Thread to finish
    ;    This ensures DllMain runs and installs the VEH *before* we resume.
    invoke  WaitForSingleObject, [h_inj_th], INFINITE
    invoke  CloseHandle, [h_inj_th]

    ; 7. Resume the Main Thread
    ;    The target now runs with your VEH active from the very first instruction.
    invoke  ResumeThread, [pi.hThread]

.cleanup:
    invoke  CloseHandle, [pi.hProcess]
    invoke  CloseHandle, [pi.hThread]

.exit:
    invoke  ExitProcess, 0

section '.idata' import data readable writeable
    library kernel32,'KERNEL32.DLL'
    include 'api\kernel32.inc'